Add cudax::distributed and cudax::returned_to specifiers for cooperative algorithms

[Jacobfaib]

Description

Add 2 more specifiers:

• cudax::distributed
• cudax::returned_to(rank)

For use in cooperative/distributed algorithms. These are in addition to cudax::broadcasted which allow the algorithm to specify how results are distributed across the participating callers.

• broadcasted: All participating ranks will receive an identical result.
• returned_to(rank): Only the specified rank will receive the result. All other ranks receive either no result or undefined values. It is up to the algorithm to define their precise value.
• distributed: All participating ranks receive some slice of the result. For example when sorting, all ranks will receive a slice of the globally sorted array. Or when doing an exclusive/inclusive scan, all ranks will receive their portion of the global scan.

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[Jacobfaib]

I wanted a common base class so that algorithms which handle multiple result policies can define the parameter as:

template <class Policy, ...>
algorithm(__result_policy_base<Policy> __res, ...)

instead of a generic T

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

The result policy utility adds returned_to and distributed, changes broadcasted_t to derive from a shared base, and replaces the removed utility header in cooperative algorithm includes. The broadcasted reduce example is marked for Doxygen and added to its example paths.

Changes

Result policy integration

Layer / File(s)	Summary
Result policy types and objects cudax/include/cuda/experimental/__utility/result_policy.cuh	The header guard changes, broadcasted_t derives from a shared base, and returned_to and distributed policy types and objects are added.
Cooperative header dependency updates cudax/include/cuda/experimental/__coop/any_of.cuh, cudax/include/cuda/experimental/__coop/reduce.cuh	Both cooperative headers replace the removed broadcasted utility include with result_policy.cuh.
Broadcasted reduce example documentation cudax/test/coop/reduce/this_block.cu, docs/cudax/Doxyfile	The broadcasted reduce call receives documentation markers, and the reduce directory is added to Doxygen’s example paths.

Suggested reviewers: ericniebler, andralex, davebayer

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: d62a5015-08a4-4069-b0ec-64ecb1bb14b1

📥 Commits

Reviewing files that changed from the base of the PR and between faa1c1c and cdd19e1.

📒 Files selected for processing (3)

• cudax/include/cuda/experimental/__coop/any_of.cuh
• cudax/include/cuda/experimental/__coop/reduce.cuh
• cudax/include/cuda/experimental/__utility/result_policy.cuh

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 47m 31s: Pass: 100%/57 | Total: 8h 11m | Max: 47m 28s | Hits: 79%/47414

See results here.